Fast Search for Multimedia Metadata in an XML Data Repository

摘要

Performance of multimedia metadata management depends on the storage, indexing, and proper schema design of the system. In this paper, we propose a design strategy for improvement of search performance to build a prototype multimedia metadata system (MMS). The proposed design strategy includes the following aspects. First, we model and design the MMS using native XML database. The native XML database stores XML documents directly without going through normalization processes required in an RDBMS. We do not store XML document on top of an RDBMS, either. Second, we introduce an assistant document, the rotated document for each searchable descriptor. Optimal inter-document schema is designed to search and locate documents in the data repository. Third, on the rotated documents, we perform lexical processing including word separating, stemming and stop word removal. Finally, different index structures are implemented and tested to build fast indices for the system. Experiments show that the proposed design achieves a speedup factor of 9.3 compared with the direct DOM method. It can speed up even more than a system based on RDBMS. It also has a better scalability.